Ling-3.0-flash-Fin: Ant Group's 127.5B-Parameter Financial Model Released on HuggingFace

On September 3, 2026, the model inclusionAI/Ling-3.0-flash-Fin was published on HuggingFace with the text-generation task. It is the first model in the Ant Ling family with enhanced financial specialization: it was fine-tuned on financial data together with specialized institutions and experts. As described by the author, it supports end-to-end financial research scenarios — from information retrieval and source verification to calculations, modeling, and report preparation.

What it means

The model has 127.5 billion parameters, and the license is MIT. The full repository of model weights in safetensors format takes up 237.5 GB, with the largest file being 5.7 GB. For running on your own hardware, the platform offers three quantized GGUF builds; the most compact working one is IQ4_XS by mradermacher, weighing 63.8 GB. This is still server-grade hardware territory: loading such a build requires a system with a corresponding amount of system RAM or multiple GPUs. The author mentions 124B total and 5.1B activated parameters with a 256K context window, indicating a MoE architecture, though these details are not disclosed in the model fact sheet. The model is not intended for local running on a typical workstation, but the open license and the availability of quantized builds leave the option of deploying it on your own server. You can follow the release of more compact versions and other news on running models on your own hardware in the CPU3D news feed.